Is the Agentic AI Market growing now?
In our agentic AI market deck, you will find everything you need to understand the market
SUMMARY
Yes, the agentic AI market is growing now. The clearest evidence is not the number of products calling themselves agents, but the simultaneous rise in production deployments, usage, recurring revenue, enterprise spending, and the infrastructure being built around them.
The headline market-size numbers are already enormous, but they are also messy. Gartner's broad estimate of $206.5 billion in AI agent software spending this year captures agentic capabilities embedded across enterprise software, so it is better read as a budget shift than as pure-play agent revenue.
The real acceleration is concentrated in large enterprises. McKinsey's latest data show agent scaling rising from 27% to 40% at companies above $1 billion in revenue, while smaller-company adoption stayed at 22%.
Production deployment is no longer unusual enough to dismiss as experimentation. Confluent found 32% of organizations already running agentic AI in production, with another 39% in pilots or early deployment, although 77% of production users had also experienced stalled projects.
The strongest platforms are not just adding customers; existing customers are pushing more work through agents. Microsoft, Salesforce and OpenAI all report sharp increases in active usage or consumption, which is a much harder growth metric to fake than launch announcements or pilot logos.
Commercial traction is now large enough to stand on its own. Agentforce has reached $1.2 billion in ARR, while Cognition, Sierra, Legora and other specialists have built businesses measured in hundreds of millions or more than $100 million of recurring or annualized revenue.
The market is taking shape first in workflows where work already happens inside software and results can be checked quickly. Coding, customer service and legal work are ahead because the agent can act on digital systems, use structured context and hand a result back to a human who can judge it.
Agentic AI is also starting to change how software is bought. Some companies are already skipping software purchases because agents let them build internally, while vendors are shifting from seat pricing toward credits, work units and outcome-based pricing.
A second market is forming around the agents themselves: identity, security, observability, testing, governance and billing. Companies do not build this control layer at scale unless the underlying deployment problem is becoming real.
The limit is autonomy, not demand. Only 5% of organizations in Grant Thornton's survey allow agents to make high-stakes decisions without human review, while data quality, reliability, governance and cost are still stopping many deployments from scaling cleanly.
The result is a market that is already commercially real but narrower than the hype suggests. Agentic AI is expanding fastest as bounded digital labor inside large organizations, not as unsupervised digital employees replacing whole functions.
So the growth case is strong even if some valuations are not. Spending, deployment, usage and revenue are rising together, while 50x to 80x revenue multiples at some leading startups show that investors are still pricing in far more future dominance than today's businesses have proved.
This market map, featured in our agentic AI market deck, highlights top companies and startups in the agentic AI market
What actually counts as agentic AI today?
Agentic AI today means software that can take a goal, work through several steps and actually do something in another system with limited human direction.
That definition excludes a lot of products now marketed as “agents.” A chatbot that answers a question is still a chatbot. A copilot that drafts an email after every human instruction is closer to an assistant. We start calling the system genuinely agentic when it can decide what to do next, use tools, retrieve information, take actions and keep working toward an outcome.
The distinction is important when measuring this market because “agentic AI” has become an extremely elastic label. Gartner has repeatedly warned about “agent washing,” where existing chatbots, assistants or automation products get renamed as agents. Its earlier review estimated that only around 130 of the thousands of vendors using agent language offered what it considered meaningful agentic capabilities.
For this analysis, we count coding agents such as Devin and Codex, customer-service agents such as Sierra and Agentforce, legal agents, recruiting agents, agents operating inside enterprise workflows, and the software needed to train, secure, monitor and govern them. Counting every AI assistant would make the market look much larger without telling us much about whether autonomous software is actually taking off.
How big is the agentic AI market right now?
The agentic AI market is already large under broad definitions, and spending is growing far faster than a normal enterprise software category.
Gartner currently forecasts roughly $206.5 billion of AI agent software spending this year, up from about $86.4 billion last year. That works out to around 2.4 times as much spending in one year.
That number needs some care. Gartner's definition captures agentic capabilities embedded across enterprise software, so $206.5 billion should not be read as revenue earned by pure agent startups. It tells us more about how quickly agents are entering software budgets than about the size of a clean standalone industry.
The narrower markets are much smaller. Gartner separately forecasts only tens of billions of dollars for some specific cross-functional agent and assistant categories later this decade. That enormous gap between definitions is why market-size reports on agentic AI can differ by an order of magnitude and still appear plausible.
The growth direction is much less ambiguous than the absolute size. Whether we measure embedded agent software, independent agent companies or enterprise usage, more money is flowing into the category now than a year ago.
If you want more recent data on this point, please see our latest agentic AI market report.
As this chart shows, and as featured in our agentic AI market deck, search interest in AI agents has been rising rapidly
Are companies actually putting AI agents into production now?
Yes, AI agents are currently running in production at enough companies that the market has clearly moved beyond the demo stage.
Confluent's latest Data Streaming Report surveyed 4,625 IT leaders across 14 countries and found that 32% of organizations now have agentic AI in production, compared with 29% a year earlier. Another 39% were in pilot or early deployment.
McKinsey's latest State of AI survey gives us a stricter picture. Among 1,719 respondents across 97 countries, 40% of people working at companies with more than $1 billion in annual revenue said their organizations were scaling agents in at least one function. The figure was 22% at smaller organizations.
The surveys define production and scaling differently, so we would not combine them into one adoption rate. What is clear is that production deployment now sits somewhere between a meaningful minority and a large minority of companies, depending on how demanding the definition is.
The uncomfortable part is what happens after launch. Confluent found that 77% of organizations already running agentic AI had experienced stalled projects. Production is becoming common enough to expose problems that never appear in a polished demo.
| Source | What was measured | Current result | Previous result |
|---|---|---|---|
| McKinsey | Large companies scaling AI agents | 40% | 27% |
| McKinsey | Smaller companies scaling AI agents | 22% | 22% |
| Confluent | Organizations with agentic AI in production | 32% | 29% |
| Confluent | Production users reporting stalled projects | 77% | — |
Is agentic AI adoption really accelerating?
Yes, agentic AI adoption is accelerating now, although the speed depends heavily on which companies we look at.
McKinsey gives us the cleanest year-over-year comparison. Among companies above $1 billion in revenue, agent scaling rose from 27% to 40%. That is a 13-point jump in a single year and roughly a 48% increase in the share of large companies scaling agents.
Usage data from individual platforms points in the same direction. Microsoft recently said monthly active usage of its own first-party agents had increased sixfold year to date. Salesforce reported 3.8 billion Agentic Work Units completed across Agentforce and Slack, with the total growing 111% quarter over quarter. ServiceNow says its agentic deployments multiplied ninefold in nine months.
A more recent OpenAI analysis makes the change visible across jobs rather than companies. Since February, weekly active enterprise Codex users increased 108 times in legal, 41 times in sales, 41 times in recruiting and 26 times in marketing. Engineering, where agentic coding started earlier, grew fivefold over the same period.
Those growth rates partly reflect small starting bases, particularly outside engineering. Still, we are seeing expansion across several unrelated platforms and functions at the same time. Agent adoption is spreading beyond the first developer-heavy wave.
This chart, included in our agentic AI market deck, illustrates yearly VC funding for agentic AI startups
Is agentic AI growth mostly a big-enterprise story?
For now, agentic AI growth is heavily concentrated in large enterprises rather than spreading evenly through the whole economy.
McKinsey's latest comparison is unusually stark. Agent scaling at companies with more than $1 billion in annual revenue jumped from 27% to 40%, while adoption at smaller organizations stayed flat at 22%.
Large companies have several advantages here. They have more repetitive knowledge work, larger support operations, more internal software, more proprietary data and enough spending to justify building the integrations around an agent. They can also tolerate a project that takes months before the economics are obvious.
Microsoft says nearly 90% of the Fortune 500 now have active agents built through its low-code and no-code tools. Sierra says more than 40% of the Fortune 50 use its customer-experience agents. Those are vendor disclosures, but the concentration fits the independent survey data.
So when people say “companies are adopting agents everywhere,” the wording is too broad. The current boom is strongest inside large enterprises and technology-heavy organizations. Smaller companies are participating, but they are not driving the acceleration yet.
If you want more recent data on this point, please see our latest agentic AI market report.
Are companies paying real money for AI agents?
Yes, companies are already spending serious money on AI agents, and several agent products have become businesses measured in hundreds of millions or more than a billion dollars.
Salesforce says Agentforce has reached $1.2 billion in annual recurring revenue, up 205% year over year. ServiceNow says its broader AI business has crossed $1 billion in annual contract value and has raised its full-year AI ACV target to $1.5 billion.
The startup numbers are now large enough to matter too. Documents reviewed by The New York Times' DealBook showed Sierra reaching about $200 million in annualized revenue in May, up more than fivefold from its previous fiscal year. Cognition reported a $492 million annualized revenue run rate for its coding business when it announced its latest funding round. Legora recently crossed $100 million in ARR in legal AI.
We should not add those figures together and call the result “market revenue.” ARR, annualized revenue and ACV are different measures, while ServiceNow's AI figure contains more than pure agent revenue.
But this is already enough to answer the commercial question. Customers are paying for agents at a scale that would make several individual products meaningful public software businesses on their own.
| Company | Latest disclosed agent-related revenue measure | Recent growth |
|---|---|---|
| Salesforce Agentforce | $1.2B ARR | +205% year over year |
| ServiceNow AI | More than $1B ACV | Agentic deployments up 9x in nine months |
| Cognition | $492M annualized revenue run rate | Enterprise Devin usage reportedly +50% month over month for six months |
| Sierra | About $200M annualized revenue | More than 5x its previous fiscal year |
| Legora | More than $100M ARR | Crossed the threshold during its latest expansion |
This chart, included in our agentic AI market deck, shows how Cognition is positioned in agentic AI
Where are AI agents actually useful today?
AI agents are working best today in coding, customer service, legal work and other digital workflows where the result can be checked quickly.
Coding remains the clearest example. McKinsey found that about one in five organizations are already scaling coding agents, rising to 31% among large enterprises. Microsoft says nearly 140,000 organizations use GitHub Copilot and enterprise subscriptions have almost tripled year over year.
The work agents handle is also becoming more substantial. Uber's CTO recently said around 10% of code produced across its roughly 8,000 engineers and technical workers is now generated autonomously. Virgin Atlantic told OpenAI that one legacy-code refactoring task that previously took two weeks could be completed with Codex in around 30 minutes.
Customer service is the second obvious market. Sierra's roughly $200 million annualized revenue comes from agents handling customer interactions, while Microsoft says nearly 60% of its service customers are already buying usage-based AI credits. Salesforce has built much of Agentforce around service workflows.
Legal AI is now moving quickly as well. Harvey and Legora have both built businesses above $100 million in recurring revenue, and Google Cloud this week introduced Gemini Enterprise for Legal with launch customers including Cleary Gottlieb, Freshfields, Weil and Williams & Connolly.
The pattern is pretty simple. Agents work best where the task already happens inside software, enough examples exist to teach the system what good work looks like, and someone can judge whether the outcome is correct.
Do companies keep using AI agents after the first deployment?
Yes, the strongest agent platforms are seeing customers use their agents more after deployment rather than leaving them as forgotten pilots.
Microsoft's sixfold increase in monthly active usage of first-party agents is one example. The company also reported that consumption of Copilot Credits nearly doubled quarter over quarter as customers extended Copilot into custom workflows.
OpenAI is seeing a similar widening inside its enterprise base. Its recent analysis found that the top 10% of enterprise customers by AI usage now generate 8.3 times as many output tokens per active user as typical firms, compared with a 2.6-times gap at the start of the year. The firms that learn how to use agents effectively appear to keep pushing much more work through them.
Salesforce's usage figures show the same kind of expansion. Its Agentic Work Units grew 111% quarter over quarter, while more than half of Agentforce and Data 360 bookings in its latest reported quarter came from existing customers.
This is one of the better tests of whether the market is real. A vendor can give away pilots and announce logos. Sustained growth in usage inside existing customers is much harder to manufacture.
This chart, included in our agentic AI market deck, illustrates yearly funding for agentic AI startups
Are AI agents already paying for themselves?
AI agents are already producing strong returns in some workflows, but plenty of deployments still cost more than companies expected.
The successful examples can be striking. HSBC says Microsoft agents reduced customer issue-resolution time by more than 30%. T-Mobile told investors in agentic marketing startup Gradial that its system cut campaign-execution time by roughly 80% to 90% while maintaining 99% accuracy. Uber says autonomous code generation has reached around 10% of its engineering output, which is meaningful at the company's scale.
Companies are also changing the way they charge because the work can increasingly be measured. Sierra sells many deployments on outcomes rather than seats or token consumption. Norm, an AI-native legal services company, also charges clients around outcomes while human lawyers supervise its agents.
Costs can rise very quickly once an agent starts looping through models, tools and other agents. A recent McKinsey analysis estimated that some customer-facing banking workflows can cost $20,000 to $30,000 to operate with a single agent and $100,000 to $200,000 with a multi-agent team. Those figures depend heavily on the workflow, but they show how quickly an exciting demo can turn into an expensive production system.
The ROI question has already split the market. Well-defined tasks with expensive human labor can work extremely well. Vague “let the agent handle everything” projects are much harder to defend economically.
If you want more recent data on this point, please see our latest agentic AI market report.
How autonomous are AI agents today?
Most AI agents today still operate with clear limits, especially when money, legal decisions or other high-stakes actions are involved.
Grant Thornton's latest AI Impact Survey found that only 5% of organizations allow agents to execute high-stakes decisions without human review. Sixty percent limit agents to moderate-risk task automation.
Recent deployments show what that looks like in practice. Fiserv's agentOS for banks was designed with policy controls, auditability and human oversight built into the system. The U.S. Army is training AI agents for cyber roles such as development, data engineering and analysis, while humans remain responsible for decisions carrying meaningful operational risk.
Even the companies building the most advanced agents talk constantly about guardrails, approvals and escalation. Sierra has been adding staged releases and approval systems around changes to live agents. OpenAI's enterprise agent products similarly emphasize approved actions and escalation to people.
“Autonomous agent” still needs some discipline as a term. Agents can already complete substantial work independently, but unrestricted autonomous decision-making remains rare inside serious organizations.
This chart, included in our agentic AI market deck, compares the main business model options for autonomous AI agent platforms
Is a new infrastructure market forming around AI agents?
Yes, agent deployment is now creating a separate market for agent security, identity, testing, observability, governance and billing.
Microsoft says tens of thousands of companies are using Agent 365 to manage tens of millions of agents. It has also formalized usage-based Copilot Credits, where agent costs depend on model tokens, tools and the work performed. ServiceNow is building an AI Control Tower to govern agents across enterprise systems.
Security companies are moving quickly. Okta recently agreed to acquire Permiso Security in a deal reported at just under $200 million, explicitly citing the need to protect AI agents and other machine identities. Prime Intellect raised $130 million to help enterprises build and train their own agents. A new startup called Arga has just raised $10 million to build realistic enterprise-software environments where agents can be trained and tested before touching live systems.
Observability is joining the stack too. Grafana Labs, which recently crossed $600 million in ARR, says companies increasingly need to monitor agent behavior alongside conventional infrastructure and has introduced dedicated Agent Observability products.
This secondary market is useful evidence. Companies generally do not spend money controlling technology they never deploy. Agent sprawl is already large enough to create problems worth paying other companies to solve.
Are AI agents starting to eat traditional SaaS budgets?
Yes, AI agents are beginning to replace some conventional software purchases, and the effect could become much larger if agents become the main interface for enterprise work.
McKinsey found that 32% of respondents said their organizations had already decided against buying at least one software product or feature because agentic coding tools allowed them to build it internally. Among the heaviest AI users, the share was even higher.
Gartner recently estimated that as much as $234 billion of enterprise application spending could be exposed to what it calls “agentic arbitrage” by 2030. That represents roughly 20% of projected enterprise SaaS spending. The basic idea is that if an agent can operate five systems for an employee, the employee may no longer need five expensive interfaces and five full seat licenses.
Software companies are already preparing for this. Microsoft increasingly combines seats with consumption credits. Salesforce is charging through Agentic Work Units and other usage-based models. Sierra built its model around successful outcomes.
We should expect agentic AI growth to show up partly as new spending and partly as money moving out of older SaaS products. That makes the eventual market larger than a simple “AI agent software revenue” number suggests, while making life uncomfortable for software companies whose main value comes from the interface humans currently click through.
If you want more recent data on this point, please see our latest agentic AI market report.
This chart, featured in our agentic AI market deck, shows the share of revenue generated by each customer segment in the agentic AI market
Are investors still piling money into agentic AI?
Yes, investors are still putting huge amounts of capital into agentic AI, with the largest checks increasingly going to companies that already have customers and revenue.
Cognition raised more than $1 billion at a $26 billion post-money valuation. Sierra raised $950 million at a $15.8 billion valuation. Harvey raised $200 million at an $11 billion valuation, while Legora raised roughly $600 million across its latest legal-AI financing and extension.
Those four financings alone add up to around $2.75 billion.
Capital is also moving into the layers around the agents. Prime Intellect raised $130 million for agent training infrastructure. New security and identity companies are attracting large rounds, while the newly funded Arga is tackling agent testing inside simulated enterprise software.
The newest development is that some winners are already coming back to investors before their previous round has had time to age. TechCrunch recently reported that Cognition was discussing another financing at a valuation of at least $40 billion, only a few months after closing at $26 billion, with investors reportedly looking toward a possible $1 billion annualized revenue run rate.
Investor appetite remains extremely strong. It has also become more selective: the largest rounds are clustering around coding, customer service, legal AI and infrastructure rather than every generic “AI employee” pitch.
Are agentic AI valuations getting ahead of the businesses?
Yes, several leading agentic AI companies are valued at multiples that assume years of extraordinary growth from here.
Sierra offers a good example. Its latest financing valued the company at $15.8 billion. Using the roughly $200 million annualized revenue figure subsequently reported by DealBook gives an implied multiple of around 79 times annualized revenue.
Cognition's $26 billion post-money valuation against its disclosed $492 million annualized revenue run rate works out to roughly 53 times. Legora's $5.6 billion valuation came shortly after it crossed $100 million in ARR, putting the rough multiple above 50 times. Harvey's $11 billion valuation compared with the roughly $190 million ARR it reported around the end of last year lands in a similar range.
These companies are growing unusually fast, so comparing them with mature SaaS companies would be misleading. Even with that growth, 50-to-80-times revenue leaves very little room for an ordinary outcome.
The market can be growing rapidly while some investments still disappoint. Investors are pricing the leaders as future owners of entire workflows, not simply as another generation of software tools.
| Company | Valuation | Revenue measure used | Rough valuation / revenue |
|---|---|---|---|
| Sierra | $15.8B | ~$200M annualized revenue | ~79x |
| Cognition | $26B post-money | $492M annualized run rate | ~53x |
| Legora | $5.6B | >$100M ARR | <56x |
| Harvey | $11B | ~$190M ARR | ~58x |
If you want more recent data on this point, please see our latest agentic AI market report.
This chart, included in our agentic AI market deck, shows how autonomous AI agent platform technology has evolved over time
What is still stopping agentic AI from scaling faster?
Data, reliability, governance and cost are currently slowing agentic AI much more than a lack of interest.
Confluent's survey found that 66% of IT leaders see data infrastructure and data quality as barriers to agentic AI. Sixty-five percent cited governance, risk and compliance, while 68% worried about model reliability and non-deterministic behavior. Among organizations already running agents in production, 77% had experienced stalled projects.
Grant Thornton found another uncomfortable gap. Nearly three in four organizations are giving agentic systems some access to their data or processes, but only one in five has actually practiced an incident-response plan for an AI failure. More than half of CIOs and CTOs also said most of their core applications are not AI-ready.
Governance can become a problem even after successful deployment. Gartner currently expects 40% of enterprises to demote or shut down some autonomous agents by 2027 after discovering governance gaps in production.
Cost is moving up the list too. McKinsey's latest work on agent economics found that complex multi-agent workflows can reach six-figure operating costs. Companies increasingly have to decide whether a task really needs the smartest model and several collaborating agents or whether a much simpler automation would do the job.
The companies that solve these problems will keep scaling. Others can easily spend a year building an expensive agent that looked brilliant during the first demo.
| Main constraint | Recent evidence | What companies are running into |
|---|---|---|
| Data | 66% cite infrastructure or quality issues in Confluent's survey | Agents act on fragmented, stale or poorly structured information |
| Reliability | 68% cite LLM reliability and non-determinism | The same workflow can behave differently across runs |
| Governance | Only 20% in Grant Thornton's survey have practiced an AI failure response | Companies are giving agents access before building the controls around them |
| Cost | Multi-agent banking workflows can reach six figures in McKinsey's analysis | More autonomy often means more model calls, tools and retries |
Are startups or big software companies winning the agentic AI market?
Neither side has won the agentic AI market yet, but the competition is getting much harder for standalone startups.
Startups proved that customers would pay first. Sierra built a roughly $200 million annualized customer-service business. Cognition reached a $492 million annualized run rate in coding. Harvey and Legora both crossed $100 million in legal AI. Those companies moved faster than the traditional enterprise platforms in their specific workflows.
The large platforms are now pushing directly into the same territory. Salesforce agreed to acquire customer-service agent company Fin. Google Cloud has just launched a dedicated legal agent platform. OpenAI recently introduced Presence for enterprises and is expanding agents through Codex and ChatGPT Work. Microsoft and ServiceNow already sit inside the identity, data and workflow systems that enterprise agents need to use.
Distribution gives the incumbents a serious advantage. A company that already stores a customer's CRM records, permissions, documents or IT workflows can attach an agent without asking the customer to rebuild the entire stack.
Startups still have room where they can own the workflow rather than merely add another agent interface. Sierra's outcome-based customer service and the legal platforms built by Harvey and Legora are examples. Their problem is that every successful vertical is now attracting the model companies and enterprise giants.
We are likely to end up with a mixed market: big platforms supplying the context and control layer, specialist companies owning valuable workflows, and frontier model providers pushing further into both.
In our agentic AI market deck, we identify pain points entrepreneurs should prioritize
Is the agentic AI market growing now?
Yes, the agentic AI market is growing fast right now, and the evidence has become much stronger than another cycle of product announcements.
The strongest evidence comes from several directions at once. Companies are putting agents into production. Existing deployments are consuming more agent work. Large enterprises are scaling adoption much faster than they were a year ago. Agent products have reached hundreds of millions and, in Salesforce's case, more than a billion dollars of recurring revenue. Investors are still writing enormous checks. At the same time, entirely new businesses are appearing around agent identity, testing, governance, observability and consumption billing.
McKinsey's latest survey shows the clearest limitation: the acceleration is concentrated in large enterprises while smaller-company adoption has barely moved. Confluent also finds that most organizations already using agents in production have run into stalled projects. Only 5% of organizations in Grant Thornton's survey are currently comfortable letting agents make high-stakes decisions without human review.
Those limits change the picture of what is growing. Today's agentic AI market is mainly about software taking over bounded pieces of digital work: writing and changing code, resolving customer issues, researching legal questions, screening candidates, operating business systems and coordinating multi-step tasks. Full digital employees running important functions without supervision remain rare.
For the market-growth question, that distinction no longer changes the answer. Spending is rising, deployments are rising, usage inside successful deployments is rising, revenue is rising, and the surrounding infrastructure market is growing with it.
Agentic AI is already a real growth market today. The hype is still ahead of the autonomy, and some valuations are far ahead of the businesses, but commercial adoption itself has crossed the line from experiment to expansion.
OUR METHODOLOGY
This analysis tests whether the agentic AI market is genuinely growing now. We broke the question into the dimensions that would have to move if a real market were forming: enterprise adoption, production deployment, usage intensity, customer spending and revenue, investment, infrastructure formation, and the economics and constraints affecting further scale.
We prioritized evidence showing what companies are actually doing. Production deployments carry more weight than pilots; expanding usage inside existing customers carries more weight than announced customer logos; recurring revenue and consumption carry more weight than product launches; and evidence appearing across several companies or surveys carries more weight than an isolated success story.
We did not combine unlike metrics into a single market score. Survey adoption rates, ARR, annualized revenue, ACV, platform consumption, funding rounds and infrastructure spending measure different things. We used each for what it can tell us directly, then looked for convergence across the different parts of the market.
We also looked for evidence that weakens or narrows the growth case. Stalled deployments, limited autonomy, data problems, governance gaps and high operating costs help show what is actually scaling today rather than what vendors hope agents will eventually become.
Recency matters heavily because the category is changing unusually quickly. We therefore prioritized recent company disclosures, earnings materials, large enterprise surveys, first-hand adoption data and research from established technology and management institutions. Vendor figures were most useful when they measured observable activity on the vendor's own platform.
Key sources include Gartner's 2026 AI agent software spending forecast, Gartner's work on agent washing and project cancellations, McKinsey's State of AI 2026, Confluent's 2026 Data Streaming Report, Microsoft's FY2026 Q3 earnings disclosure, Salesforce's FY2027 Q1 results, Salesforce's Agentic Work Units methodology, OpenAI's Enterprise Signals, OpenAI's enterprise adoption analysis, and Grant Thornton's 2026 AI Impact Survey.
For agent economics and software-budget displacement, we also used McKinsey's analysis of agentic workflow economics and Gartner's estimate of enterprise software spending exposed to agentic arbitrage. For the emerging control layer, we used Microsoft's Agent 365 disclosure, Okta's Permiso Security acquisition announcement, Grafana Labs on AI-era observability, and Google Cloud's Gemini Enterprise for Legal launch.
The final conclusion does not depend on one market-size estimate. It comes from the overlap between deployment, usage, revenue, customer spending, investment and the supporting infrastructure market, while keeping the current limits on autonomy and economics in view.
This chart, included in our agentic AI market deck, shows the share of revenue by region across Europe, Asia, North America, Africa, and South America in the agentic AI market
